Direct access to WAIC2022丨Jiuzhang Yunji DataCanvas Company made a wonderful appearance with its causal learning technology achievements

The 2022 World Artificial Intelligence Conference (WAIC) with the theme of "Intelligent Connected World, Unbounded Life" concluded successfully in Shanghai on September 3 . WAIC, as a "technological vane, application showcase, industry accelerator, and governance forum" for global artificial intelligence, is the most influential industry event in the global artificial intelligence field.

"WAIC 2022 · AI Developers Day" is one of the most important technical forums of the WAIC conference. With the theme of "What AI developers really care about", it brings together the 2021 Turing Award winners and Chinese and foreign academicians. , world-class technical experts and technology company founders, and 15 key guests from academia and industry. Yang Jian, Vice President of Open Source Technology and Director of D-Lab of Jiuzhang Yunji DataCanvas Company, was invited to attend the forum and focused on how to use a complete, comprehensive, end-to-end causal learning toolkit to solve "causal discovery and causal quantity identification" ", causal effect estimation, counterfactual inference and strategic learning" five key issues, and delivered a wonderful keynote speech "YLearn: Causal Learning, From Prediction to Decision".

Causal learning: a technological breakthrough in the development of artificial intelligence

As machine learning and deep learning encounter technical bottlenecks in the development process, artificial intelligence The speed of intelligence development is gradually slowing down. The reasons are, on the one hand, that machine learning has key problems such as weak generalization ability, weak interpretability, and insufficient decision support capabilities; on the other hand, governments and enterprises have proposed the concept of "intelligent decision-making" The demand is to realize automated decision-making in a data-driven manner to improve overall operational efficiency.

With the increasing application of machine learning modeling, artificial intelligence technology has been upgraded from predictive analysis to guided analysis, and automated "decision-making" has become the core need of governments and enterprises in the era of digital intelligence. Decision makers need an understandable AI decision-making logic and credible and explainable decision-making results. However, current machine learning is mainly used to complete predictive tasks, which is difficult to meet the needs of governments and enterprises for automated decision-making.

The "2022 Emerging Technology Hype Cycle" released by Gartner mentioned that causal artificial intelligence is one of the key technologies to accelerate AI automation. Causal learning has become a key technology to supplement machine learning problems. It is a technological breakthrough with great potential for the development of artificial intelligence, which has attracted widespread attention and hot research in the industry.

YLearn: Causal learning, from prediction to decision-making

Mr. Yoshua Bengio, the winner of the 2019 Turing Award, once mentioned that "causality is very important for the next progress of machine learning." Since 2019, new academic research results on causal learning have continued to appear, and the number of related papers published has doubled every year. At present, judging from the research and development of causal learning at home and abroad, there are many causal learning tools, such as DoWhy, EconML that focuses on solving causal effect evaluation problems, CausalML that is used to complete uplift modeling, and CausalLearn that focuses on solving causal discovery problems. . However, these tools can only solve some of the problems in causal learning, and because different tools rely on different theoretical frameworks and structural systems, it is difficult to integrate and use the tool packages. The field of causal learning lacks a systematic, complete, comprehensive, and end-to-end tool kit.

The one-stop open source algorithm toolkit YLearn independently developed by Jiuzhang Yunji DataCanvas Company handles the complete process of causal learning. It is currently the first end-to-end, more complete, The more systematic causal learning algorithm toolkit takes the lead in solving the five key problems of "causal discovery, causal quantity identification, causal effect estimation, counterfactual inference and strategy learning" in causal learning, reducing the number of "decision makers"Use threshold to continuously meet the needs of governments and enterprises for automated "decision-making".

​

​Click to get the GitHub address​​

YLearn consists of CausalDiscovery, CausalModel, EstimatorModel, Policy, Interpreter, Whatif and other components Composed, each component supports independent use and unified packaging. To help users understand data and adjust strategies more intuitively, YLearn provides visual output of important modules such as causal diagrams, causal effect explanations, and decision trees.

Compared with domestic and foreign cause-and-effect learning tools, YLearn of Jiuzhang Yunji DataCanvas Company has the characteristics of

one-stop, new and comprehensive, and wide-ranging .

•

One-stop shop

The usual causal learning process includes discovering the causal structure from the data, building a causal model for the causal structure, using the causal model to identify the causal effect and estimating the causal effect from the data. YLearn supports these functions in one stop, allowing users to use and deploy causal learning at the lowest learning cost.

• New and Complete

YLearn implements a number of various algorithms developed in the field of causal learning in recent years, such as Meta-Learner, Double Machine Learning, etc. We will also be committed to keeping up with cutting-edge progress and keeping causal identification and estimation models advanced and comprehensive.

• Wide use

YLearn supports interpretation of estimated causal effects, selecting the most profitable option among various options based on causal effects, and visualizing the decision-making process, etc. Function. In addition, YLearn also supports small functions such as outputting the probability distribution expression of the causal effect identified in the causal structure in the form of LaTex, helping users intersect causal learning with other directions.

Combined with the needs of governments and enterprises in decision-making tasks, YLearn will be combined with the automatic machine learning platform of Jiuzhang Yunji DataCanvas Company. Through the integration of AutoML technology, Improve the robustness, generalization ability and interpretability of machine learning, realize automatic parameter adjustment and optimization of causal learning, and further reduce the threshold for use. At the same time, YLearn solves the "stuck" problem of the lack of a powerful and complete causal learning toolkit in the market, returns technology to business, supports decision-making business scenarios, and provides customers with a variety of decision-making solutions.

Causal learning helps artificial intelligence move into a new stage

As the core force of a new round of technological revolution and industrial transformation, artificial intelligence technology is in a new stage of development from prediction to decision-making. Causal learning plays an important role at this stage, making up for the theoretical flaws of machine learning, gradually solving the problem from "what" to "why", and improving the credibility and credibility of "AI decision-making" based on the needs of governments and enterprises. Availability, further handing over AI capabilities to business use.

In order to better promote the development of the domestic field of causal learning and promote the diversified development of causal learning, Jiuzhang Yunji DataCanvas Company jointly joined the World Artificial Intelligence Conference Organizing Committee Office, Machine Heart, Shanghai The Municipal Artificial Intelligence Industry Association and Tianchi jointly organized the Hackathon "Causal Learning and Decision Optimization Challenge" to provide a platform for elite developers from all over the world to compete on the same stage. With the theme of "How to optimize intervention plans to maximize causal effects", the challenge concretizes universal issues in causal learning and aims to test contestants' estimation abilities in decision-making using causal inference.

As the industry's first competition for the "whole process of causal inference", it has received awards from all over the country, including companies that use artificial intelligence-related technologies to empower digital upgrades, and combine artificial intelligence technology Nearly 4,000 teams, including scientific research units, teams from colleges and universities, and professional developers engaged in innovative exploration, signed up to participate. After 23 days of competition on the same stage, the participating teams continued to explore the technical peaks in the field of causal learning, set new performance records, and compete to select the TOP18 winning teams with strong AI technical strength and creativity.

In the future, Jiuzhang Yunji DataCanvas will continue to innovate and develop open source tools, combine the business needs of governments and enterprises with technical practices, and help governments and enterprises become digitally intelligent. Upgrade and promote artificial intelligence to a new stage.

The above is the detailed content of Direct access to WAIC2022丨Jiuzhang Yunji DataCanvas Company made a wonderful appearance with its causal learning technology achievements. For more information, please follow other related articles on the PHP Chinese website!